update

app.py

@@ -1,63 +1,180 @@
-import gradio as gr
-from huggingface_hub import InferenceClient
-
 """
-For more information on `huggingface_hub` Inference API support, please check the docs: https://huggingface.co/docs/huggingface_hub/v0.22.2/en/guides/inference
+来自 https://github.com/OpenLMLab/MOSS/blob/main/moss_web_demo_gradio.py
+
+
+# 单卡报错
+python moss_web_demo_gradio.py --model_name fnlp/moss-moon-003-sft --gpu 0,1,2,3
+
+# TODO
+- 第一句：
+- 代码和表格的预览
+- 可编辑chatbot：https://github.com/gradio-app/gradio/issues/4444
 """
-client = InferenceClient("HuggingFaceH4/zephyr-7b-beta")
+
+from transformers.generation.utils import logger
+
+import gradio as gr
+import argparse
+import warnings
+import torch
+import os
+# from moss_util import generate_query
+from models.qwen2_util import bot
+# generate_query = None
+
+# gr.ChatInterface
+
+# from gpt35 import build_message_for_gpt35, send_one_query
+
+#
+# def postprocess(self, y):
+#     if y is None:
+#         return []
+#     for i, (message, response) in enumerate(y):
+#         y[i] = (
+#             None if message is None else mdtex2html.convert((message)),
+#             None if response is None else mdtex2html.convert(response),
+#         )
+#     return y
+#
+#
+# gr.Chatbot.postprocess = postprocess
+
+
+def parse_text(text):
+    """copy from https://github.com/GaiZhenbiao/ChuanhuChatGPT/"""
+    lines = text.split("\n")
+    lines = [line for line in lines if line != ""]
+    count = 0
+    for i, line in enumerate(lines):
+        if "```" in line:
+            count += 1
+            items = line.split('`')
+            if count % 2 == 1:
+                lines[i] = f'<pre><code class="language-{items[-1]}">'
+            else:
+                lines[i] = f'<br></code></pre>'
+        else:
+            if i > 0:
+                if count % 2 == 1:
+                    line = line.replace("`", "\`")
+                    line = line.replace("<", "&lt;")
+                    line = line.replace(">", "&gt;")
+                    line = line.replace(" ", "&nbsp;")
+                    line = line.replace("*", "&ast;")
+                    line = line.replace("_", "&lowbar;")
+                    line = line.replace("-", "&#45;")
+                    line = line.replace(".", "&#46;")
+                    line = line.replace("!", "&#33;")
+                    line = line.replace("(", "&#40;")
+                    line = line.replace(")", "&#41;")
+                    line = line.replace("$", "&#36;")
+                lines[i] = "<br>" + line
+    text = "".join(lines)
+    return text
 
 
-def respond(
-    message,
-    history: list[tuple[str, str]],
-    system_message,
-    max_tokens,
-    temperature,
-    top_p,
-):
-    messages = [{"role": "system", "content": system_message}]
+def generate_query(chatbot, history):
+    if history and history[-1][1] is None:  # 该生成response了
+        return None, chatbot, history
+    query = bot.generate_query(history)
+    # chatbot.append((query, ""))
+    chatbot.append((query, None))
+    history = history + [(query, None)]
+    return query, chatbot, history
 
-    for val in history:
-        if val[0]:
-            messages.append({"role": "user", "content": val[0]})
-        if val[1]:
-            messages.append({"role": "assistant", "content": val[1]})
+def generate_response(query, chatbot, history):
+    """
+    自动模式下：query is None，或者 query = history[-1][0]
+    人工模式下：query 是任意值
+    :param query:
+    :param chatbot:
+    :param history:
+    :return:
+    """
+    # messages = build_message_for_gpt35(query, history)
+    # response, success = send_one_query(query, messages, model="gpt-35-turbo")
+    # response = response["choices"][0]["message"]["content"]
 
-    messages.append({"role": "user", "content": message})
+    #
+    if history[-1][1] is not None or chatbot[-1][1] is not None:
+        return chatbot, history
 
-    response = ""
+    if query is None:
+        query = history[-1][0]
+    response = bot.generate_response(query, history[:-1])
+    # chatbot.append((query, response))
+    history[-1] = (query, response)
+    chatbot[-1] = (query, response)
+    print(f"chatbot is {chatbot}")
+    print(f"history is {history}")
+    return chatbot, history
 
-    for message in client.chat_completion(
-        messages,
-        max_tokens=max_tokens,
-        stream=True,
-        temperature=temperature,
-        top_p=top_p,
-    ):
-        token = message.choices[0].delta.content
 
-        response += token
-        yield response
+def reset_user_input():
+    return gr.update(value='')
+
+
+def reset_state():
+    return [], []
+
 
 """
-For information on how to customize the ChatInterface, peruse the gradio docs: https://www.gradio.app/docs/chatinterface
+TODO: 使用说明
+
+avatar_images
 """
-demo = gr.ChatInterface(
-    respond,
-    additional_inputs=[
-        gr.Textbox(value="You are a friendly Chatbot.", label="System message"),
-        gr.Slider(minimum=1, maximum=2048, value=512, step=1, label="Max new tokens"),
-        gr.Slider(minimum=0.1, maximum=4.0, value=0.7, step=0.1, label="Temperature"),
-        gr.Slider(
-            minimum=0.1,
-            maximum=1.0,
-            value=0.95,
-            step=0.05,
-            label="Top-p (nucleus sampling)",
-        ),
-    ],
-)
-
-
-if __name__ == "__main__":
-    demo.launch()
+with gr.Blocks() as demo:
+    gr.HTML("""<h1 align="center">欢迎使用 self chat 人工智能助手！</h1>""")
+
+    gr.Textbox(value="You are a friendly Chatbot.", label="System message"),
+    system = gr.Textbox(show_label=False, placeholder="You are a helpful assistant.")
+    chatbot = gr.Chatbot(avatar_images=("assets/profile.png", "assets/bot.png"))
+    with gr.Row():
+        with gr.Column(scale=4):
+            user_input = gr.Textbox(show_label=False, placeholder="Input...", lines=10)
+            with gr.Row():
+                generate_query_btn = gr.Button("生成问题")
+                regen_btn = gr.Button("🤔️ Regenerate (重试)")
+                submit_btn = gr.Button("生成回复", variant="primary")
+                stop_btn = gr.Button("停止生成", variant="primary")
+                empty_btn = gr.Button("🧹 Clear History (清除历史)")
+        with gr.Column(scale=1):
+            # generate_query_btn = gr.Button("Generate First Query")
+
+            clear_btn = gr.Button("重置")
+            gr.Dropdown(
+                ["moss", "chatglm-2", "chatpdf"],
+                value="moss",
+                label="问题生成器",
+                # info="Will add more animals later!"
+            ),
+            gr.Dropdown(
+                ["moss", "chatglm-2", "gpt3.5-turbo"],
+                value="gpt3.5-turbo",
+                label="回复生成器",
+                # info="Will add more animals later!"
+            ),
+
+    history = gr.State([])  # (message, bot_message)
+
+    submit_btn.click(generate_response, [user_input, chatbot, history], [chatbot, history],
+                    show_progress=True)
+    # submit_btn.click(reset_user_input, [], [user_input])
+
+    clear_btn.click(reset_state, outputs=[chatbot, history], show_progress=True)
+
+    generate_query_btn.click(generate_query, [chatbot, history], outputs=[user_input, chatbot, history], show_progress=True)
+
+    gr.Slider(minimum=1, maximum=2048, value=512, step=1, label="Max new tokens"),
+    gr.Slider(minimum=0.1, maximum=4.0, value=0.7, step=0.1, label="Temperature"),
+    gr.Slider(
+        minimum=0.1,
+        maximum=1.0,
+        value=0.95,
+        step=0.05,
+        label="Top-p (nucleus sampling)",
+    ),
+
+demo.queue().launch(share=False)
+# demo.queue().launch(share=True)

models/qwen2_util.py

@@ -0,0 +1,85 @@
+"Qwen/Qwen2-0.5B-Instruct"
+
+from threading import Thread
+from simulator import Simulator
+
+from transformers import TextIteratorStreamer
+
+
+class Qwen2Simulator(Simulator):
+
+    def generate_query(self, history):
+
+        inputs = ""
+        if history:
+            messages = []
+            for query, response in history:
+                messages += [
+                    {"role": "user", "content": query},
+                    {"role": "assistant", "content": response},
+                ]
+
+            inputs += self.tokenizer.apply_chat_template(
+                messages,
+                tokenize=False,
+                add_generation_prompt=False,
+            )
+        inputs = inputs + "<|im_start|>user\n"
+        input_ids = self.tokenizer.encode(inputs, return_tensors="pt").to(self.model.device)
+        return self._generate(input_ids)
+        # for new_text in self._stream_generate(input_ids):
+        #     yield new_text
+
+    def generate_response(self, query, history):
+        messages = []
+        for _query, _response in history:
+            if _response is None:
+                pass
+            messages += [
+                {"role": "user", "content": _query},
+                {"role": "assistant", "content": _response},
+            ]
+        messages.append({"role": "user", "content": query})
+
+        input_ids = self.tokenizer.apply_chat_template(
+            messages,
+            tokenize=True,
+            return_tensors="pt",
+            add_generation_prompt=True
+        ).to(self.model.device)
+        return self._generate(input_ids)
+        # for new_text in self._stream_generate(input_ids):
+        #     yield new_text
+
+    def _generate(self, input_ids):
+
+        input_ids_length = input_ids.shape[-1]
+        response = self.model.generate(input_ids=input_ids, **self.generation_kwargs)
+        return self.tokenizer.decode(response[0][input_ids_length:], skip_special_tokens=True)
+
+    def _stream_generate(self, input_ids):
+        streamer = TextIteratorStreamer(tokenizer=self.tokenizer, skip_prompt=True, timeout=60.0,
+                                        skip_special_tokens=True)
+
+        stream_generation_kwargs = dict(
+            input_ids=input_ids,
+            streamer=streamer
+        ).update(self.generation_kwargs)
+        thread = Thread(target=self.model.generate, kwargs=stream_generation_kwargs)
+        thread.start()
+
+        for new_text in streamer:
+            yield new_text
+
+
+bot = Qwen2Simulator(r"E:\data_model\Qwen2-0.5B-Instruct")
+# bot = Qwen2Simulator("Qwen/Qwen2-0.5B-Instruct")
+
+
+#
+# history = [["hi, what your name", "rhino"]]
+# generated_query = bot.generate_query(history)
+# for char in generated_query:
+#     print(char)
+#
+# bot.generate_response("1+2*3=", history)

requirements.txt

@@ -1 +1,4 @@
-huggingface_hub==0.22.2
+huggingface_hub==0.22.2
+transformers
+torch
+accelerate

simulator.py

@@ -0,0 +1,47 @@
+from transformers import AutoModelForCausalLM, AutoTokenizer
+
+
+class Simulator:
+
+    def __init__(self, model_name_or_path):
+        """
+        在传递 device_map 时，low_cpu_mem_usage 会自动设置为 True
+        """
+
+        self.tokenizer = AutoTokenizer.from_pretrained(model_name_or_path)
+        self.model = AutoModelForCausalLM.from_pretrained(
+            model_name_or_path,
+            torch_dtype="auto",
+            device_map="auto"
+        )
+        self.model.eval()
+
+        self.generation_kwargs = dict(
+            do_sample=False,
+            temperature=0.7,
+            max_length=500,
+            max_new_tokens=10
+        )
+
+        generation_kwargs = dict(
+
+            max_length=500,
+            max_new_tokens=200
+        )
+
+        print(1)
+
+    def generate_query(self, history):
+        """ user simulator
+        :param history:
+        :return:
+        """
+        raise NotImplementedError
+
+    def generate_response(self, input, history):
+        """ assistant simulator
+        :param input:
+        :param history:
+        :return:
+        """
+        raise NotImplementedError